Geophys. Res. Led 16, 703 (1989). atmospheric composition" (NASA Conf. Publ. versity, Mainz (1990).

71 . D. Kley et a/.,in Globaland Regional Environmen- 3023, NASA, Washington, DC, 1989). 83. P.J. Crutzen and P. H. Zimmermann, TellusSer. B
tal Atmospheric Chemistry: Proceedings of the 77. S. Madronich and C. Granier, Geophys.Res. Lett. 43, 136 (1991).
International Conference on Global and Regional 19, 465 (1992). 84. The following people have been generous with
Environmental Atmospheric Chemistry (Depart- 78. A. M. Thompson, J. Geophys. Res. 89, 1341 reportsof recentwork: C. Bruhl, J. Fishman,I. S. A.
ment of Energy Conference 890525, 1989, avail- (1984); S. Madronich, ibid. 92, 9740 (1987). Isaksen, M. Kanakidou, M. A. K. Khalil, K. Law, Y.
able from the NationalTechnical InformationSer- 79. S. A. Penkett and K. A. Brice, ibid. 319, 655 Lu, A. Neftel, J. Pinto, M. Prather, and R. Prinn. M.
vice, Springfield, VA), p. 9. (1986). Kanakidou, J. Logan, and W. Stockwell made
72. D. Rind, R. Suozzo, N. K. Balanchandran, M. J. 80. J. A. Logan, J. Geophys. Res. 90, 10463 (1985). helpful comments on the manuscript. Some of the
Prather, J. Atmos. Sci. 47, 475 (1990). An update for the North American data is in J. A. model comparisons (Tables 1 and 2) grew out of
73. K. 6. Hogan, J. S. Hoffman, A. M. Thompson, Logan, in Tropospheric Ozone, I. S. A. Isaksen, the United Nations Environmental Programme-
Nature 354, 181 (1991); A. M. Thompson, K. 6. Ed. (North Atlantic Treaty OrganizationAdvanced World MeteorologicalOrganizationOzone Assess-
Hogan, J. S. Hoffman, Atmos. Environ., in press. Study Institutes, Reidel, Dordrecht, Holland, ment Group Workshop (London, June 1991) and a
74. C. S. Atherton and J. E. Penner, J. Geophys. Res. 1988), pp. 3272.44. North Atlantic Treaty OrganizationAdvanced Re-
95, 14027 (1990). 81. Locations of sites: Barrow (7I0N, 157"W); Mauna search Workshop on Methane (Mt. Hood, Oregon,
75. 1. S. A. Isaksen,T. Bemtsen,S. Solberg, in Ozone in Loa (19.5"N, 156"W);Samoa (14"S, 17I0W);South October 1991).My researchon oxidizing changes
the Atmosphere, R. D. Bojkov and P. Fabian, Eds. Pole (90°S, 0). See (36); S. J. Oltmans and W. D. is supported by NASA Programs in Earth Observ-
(A. Deepak, Hampton,VA, 1989), pp. 576.579. Komhyr, J. Geophys. Res. 91, 5229 (1986). ing Systems and Tropospheric Chemistry and by
76. M. J. Prather, Ed., "An assessment model for 82. K. M. Valentin, thesis, Johannes-Gutenburg Uni- the U.S. Environmental Protection Agency.

Molecular Epidemiology of HIV Epidemiologic lnvestigation

In July 1990, we reported that a young
Transmission in a Dental Practice woman with AIDS (patient A) had most
likely acquired her HIV-1 infection while
undergoing invasive dental procedures by a
Chin-Yih Ou, Carol A. Ciesielski, Gerald Myers, Florida dentist with AIDS (4). Following
Claudiu I. Bandea, Chi-Cheng Luo, Bette T. M. Korber, publication of the report, the dentist pub-
James I. Mullins, Gerald Schochetman, Ruth L. Berkelman, licly requested that his former patients be
tested for HIV infection. Among approxi-
A. Nikki Economou, John J. Witte, Lawrence J. Furman, mately 1I00 persons whose blood was tested
Glen A. Satten, Kersti A. Maclnnes, James W. Curran, by the Florida Department of Health and
Harold W. Jaffe, Laboratory lnvestigation Group,* Rehabilitative Services (HRS), two pa-
Epidemiologic lnvestigation Group* tients (patients B and C) were found to be
HIV-positive. An additional infected pa-
Human immunodeficiency virus type 1 (HIV-1) transmission from infected patients to tient (patient D) was ascertained by HRS
health-careworkers has been well documented, but transmissionfrom an infected health- through cross matching a list of the dentist's
care worker to a patient has not been reported.After identification of an acquired immu- former patients with the Florida AIDS case
nodeficiency syndrome (AIDS) patient who had no known risk factors for HIV infection but registry. Two other patients of the dentist
who had undergone an invasive procedure performed by a dentist with AIDS, six other (patients E and G) contacted the Centers
patients of this dentist were found to be HIV-infected. Molecular biologic studies were for Disease Control (CDC) to report that
conducted to complement the epidemiologic investigation. Portions of the HIV proviral they were HIV-infected. A former sex part-
envelope gene from each of the seven patients,the dentist, and 35 HIV-infectedpersons ner named by patient E was found to be
from the local geographic area were amplified by polymerase chain reaction and se- HIV-infected and had also been a patient of
quenced. Three separate comparative genetic analyses-genetic distance measure- the dentist (patient F). Characteristics of
ments, phylogenetictree analysis, and amino acid signaturepatternanalysis--showed that these seven infected patients and the den-
the viruses from the dentist and five dental patients were closely related. These data, tist are included in Table 1.
together with the epidemiologic investigation,indicatedthat these patientsbecameinfected Patient D had previously been reported
with HIV while receiving care from a dentist with AIDS.
C.-Y. Ou, C. A. Ciesielski, C. I. Bandea, C.-C. Luo, G.
Schochetman, R. L. Berkelman, G. A. Satten, J. W.
Curran, and H. W. Jaffe are in Division of HIV/AIDS,
Increasingly, molecular biologic tech- high degree of genetic relatedness may im- National Center for Infectious Diseases, Centers for
Disease Control,Atlanta, GA30333. G. Myers, 6. T. M.
niques have been used to study the epide- ply an epidemiologic linkage between per- Korber, and K. A. Maclnnes are in the Theoretical
miology of infectious diseases. For viral sons infected with these strains. Division, Los Alamos National Laboratory, Los
infections of humans, techniques to analyze The human immunodeficiency virus Alamos, NM 87545. J. I. Mullinsis in the Departmentof
Microbiology and Immunology, Stanford University
viral genetic sequence information, such as (HIV) has a high mutation rate (2, 3), such School of Medicine, Stanford, CA 94305. A. N. Econ-
oligonucleotide fingerprinting of RNA ge- that HIVs from different individualsare found omou and J. J. Witte are in the Florida Department of
nomes with ribonuclease, mapping of DNA to be genetically distinct (3). In this article, Health and Rehabilitative Services, Tallahassee, FL
32399. L. J. Furman is in the Division of Oral Health,
genomes with restriction endonucleases, we describe the use of genomic sequencing to National Center for Prevention Services, Centers for
and genomic sequencing, have been used to investigate a cluster of HIV infections in a Disease Control, Atlanta, GA 30333.
study viral transmissions from person to Florida dental practice. The high degree of *J. Moore, Y. Villamarzo, and C. Schable, Division of
person, within communities, and between genetic relatedness observed among the HIV HIVIAIDS and E. G. Shpaer, Departmentof Microbiol-
countries (1). Requisite to such studies is strains from a dentist with acquired immuno- ogy and Immunology, Stanford University School of
Medicine.
the existence of viral genetic variation; the deficiency syndrome (AIDS) and five of his tT. Liberti and S. Lieb, Florida Department of Health
greater the variation, the greater the power infected patients supports the epidemiologic and RehabilitativeServices (HRS); R. Scott, J. Howell,
of the methods to distinguish strains of the investigation that indicated that these pa- R. Dumbaugh, A. Lasch, Florida HRS District 9; B.
Kroesen and L. Ryan, Martin County Public Health
virus. For a virus with substantial genomic tients became infected with HIV while receiv- Unit, Florida HRS; K. Bell, V. Munn, D. Marianos, and
variation, identification of strains with a ing dental care. B. Gooch, Centers for Disease Control.

SCIENCE VOL. 256 22 MAY 1992

 to HRS as having behavioral risks for HIV (6). Several factors, including duration of were collected between March 1990 and
infection. To establish any risk factors for infection (7), host immune pressure (8), April 1991. Local control specimens were
exposure to HIV for the remaining six disease stage (9), and therapy (lo), may collected from two HIV clinics located
patients, follow-up investigations were con- contribute to the degree and rate of HIV within 90 miles of the dental practice. Of
ducted (5). These investigations included genetic variation. HIV strains from persons the 35 LCs, 7 did not have symptoms
interviews with the patients and their fam- with a known common infection link, for related to their HIV infection, 11 were
ilies and friends; review of medical, dental, example, sex partners (I I), mothers and symptomatic but had not developed AIDS,
and health de~artmentrecords: interviews their infants (11, 12), blood donors and and 17 had AIDS.
with health-care providers; and testing of recipients (13), and hemophilic men re- DNA fragments of approximately 680
the watients' sex wartners for HIV infection. ceiving common lots of contaminated base pairs containing the C2, V3, V4, C3,
~ntehiewsand Aedical records established blood products (2, I+), have been shown to and V5 domains of the gp120 gene were
that patient F also had behavioral risk exhibit a closer genetic relatedness than amplified by a two-step polymerase chain
factors for HIV infection. The other five HIV strains from persons without a direct reaction (PCR) procedure (15, 16) on DNA
patients (patients A, B, C, E, and G) transmission link. Thus, we used the degree from peripheral blood mononuclear cells
denied injecting drug use since 1978 and of genetic similarity, in combination with (PBMCs) of the dentist and the seven pa-
had no history of transfusion or receipt of epidemiologicinformation, to evaluate pos- tients. The resulting amplified DNA prod-
blood products; the two male patients (pa- sible HIV transmission linkage. ucts were sequenced directly or cloned into
tients C and G) denied having had sex with Analysis of HIV-1 genetic variation in- an M13 vector and sequenced (15). Care
men. The possibility that patient C had volves comparison of multiple sequences was taken to prevent DNA carryover during
engaged in high-risk behaviors was raised covering a region of the viral genome. the PCR procedure (15, 17), a problem that
during the epidemiologic investigation; Among the structural genes of HIV, the has been previously. reported (18, 19). We
however, behavioral exposures to HIV envelope gene (env) exhibits the highest also took measures to facilitate the manage-
could not be documented. degree of genetic diversity. The distribution ment and identification of the source of
Only one of the known sex partners of of genetic diversity in env is not uniform as specimens (15). The initial results were ver-
these five patients tested positive for HIV evidenced by the presence of interspersed ified by repeating the PCR amplification
infection. This person (patient F) had been conserved (C) and variable (V) domains in procedure with a second vial of PBMCs from
an infrequent sex partner of patient E from the external glycoprotein gp120. We fo- either the first blood collection (the dentist
1987 until the fall of 1988. Patient E was cused most of our analyses on the C2 and and patients D and. E) or a second blood
first tested for HIV infection in October V3 domains because (i) these domains con- collection (patients A, B, C, F, and G);
1988 and was seropositive; patient F had tain nucleotide sequences with sufficient products from the second round of amplifi-
been tested in October and December 1988 variability to distinguish between strains, a cation were sequenced directly (15). In ad-
and was seronegative. In September 1989, feature essential for establishing HIV trans- dition, HIV sequences from patients A and
approximately 1 year after his last sexual mission linkage, (ii) C2-V3 sequences have B and one LC were independently verified
contact with patient E and his last reported been used in previous studies of epidemio- by another laboratory (20).
dental visit, patient F had an illness com- logically linked infections (2, 11, 12, 14),
patible with an acute retroviral infection; and (iii) the HIV Sequence Database (3) The Dental Group
he first tested positive for HIV in December contains a relative abundance of C2-V3
1990. Thus, patients E and F appeared to sequences for comparative purposes. To assess the genetic relatedness between
have contracted their HIV infections from Blood specimens from the dentist, pa- the HIV strains infecting the dentist and
different sources. tients A through G, and 35 local HIV- his seven patients, multiple C2-V3 se-
The dentist was diagnosed with symp- seropositive persons (local controls or LCs) quences (Table 1) as well as V4-C3-V5
tomatic HIV infection in late 1986 and
with AIDS in Se~tember1987. when a
biopsy of his showed ~ a ~ g s isarco-
's Table 1. Dental cohort clinical information and HIV nucleotide variation in the C2-V3 domain of the
envelope gene.
ma and his CD4+ lymphocyte count was
less than 200 per microliter. He continued
Known M I 3 lntraperson lnterperson variationt (%)
to practice general dentistry for approxi- Person Sex risk Clinical status* clones variationt
mately 2 years after he was diagnosed with factor (no.) To dentist To 30 LCs*
AIDS. During this 2-year period, each of
the five patients without confirmed expo- Dentist M Yes AIDS 6 3.3 (0.8-5.4) 11.O (5.8-1 6.0)
sures to HIV made multiple visits to the Patient A F No AIDS 6 2.0 (0.0-4.5) 3.4 (0.8-6.2) 10.9 (5.4-14.8)
Patient B F No Asymptomatic 12 1.9 (0.4-3.7) 4.4 (2.1-7.0) 11.2 (6.2-16.5)
dental office for invasive urocedures. in- (CD4 = 2221~1)
cluding extractions or root canal therapy. Patient C M No§ Asymptomatic 5 1.2 (0.4-1.6) 3.4 (2.1-4.9) 11.1 (7.C15.6)
(CD4 = <50/kl)
HIV Genetic Variation Patient E F No Asymptomatic 6 2.1 (0.4-3.7) 3.4 (1.2-6.6) 10.8 (5.8-14.8)
(CD4 = 567IpI)
Because the epidemiologic findings suggest- Patient G M No Asymptomatic 5 2.8 (1.6-3.7) 4.9 (2.9-7.0) 11.8 (6.2-16.9)
(CD4 = 400lpI)
ed that transmission mav have occurred in Patient D M Yes AIDS 5 7.5 (0.0-9.9) 13.6 (11.5-15.6) 13.1 (7.8-17.3)
the dental practice, we conducted a labora- Patient F M Yes Asymptomatic 6 3.0 (0.8-5.8) 10.7 (8.2-13.6) 11.9 (7.0-17.3)
tory investigation to determine whether (CD4 = 2531~1)
there was evidence of genetic similarity
*Clinicalstatus at the time of specimen collection. CD4 is the CD4+ lymphocyte count ?Sequence alignment
among the viruses infecting the dentist and over 243 nucleotide positions in C2-V3 was obtained with the program PlMA (37)followed by manual refinement
his patients. HIV is known to undergo with MASE (3).Subsequently, all MI3 clone sequences were compared to each other and to direct sequences
genetic change or variation during its life (15) of PCR-amplifiedproducts using the SIMILARITY function of MASE. Only single nucleotide differences are
scored; gaps were not scored. The average percentages of differences are shown with the range of values in
cycle resulting in a myriad of related viral parentheses. *Sequences from five of the LCswere shorter than 243 nucleotides and were not included in this
progenies (commonly called quasi-species) analysis. §Possible risk factors for HIV infection were suggested but not documented.

1166 SCIENCE VOL. 256 22 MAY 1992

sequences from the dentist and the seven sequences were not available, were deter- that this cluster of related viruses in the
dental patients were compared. Five to 12 mined from their respective M13 clones. dentist and patients A, B, C, E, and G is
C2-V3 sequences (containing a portion of Thirty of the 35 C2-V3 sequences (both not due to chance (23). This cluster cannot
the C2 and all of the V3 region) and several direct and clone) from the LCs were longer be explained by the existence of a stable
V4-C3-V5 sequences from independent than 243 nucleotides and were included in viral variant within the geographic area.
M13 clones were obtained from each of the the genetic distance comparison (Table 1). Other data from the V4-C3-V5 region are
seven dental patients. We defined the av- In calculating the genetic distances in Ta- consistent with our analysis of the C2-V3
erage or genetic distance between the virus- ble 1, we used direct sequences only when region sequence relationships.
es of one individual and those of another clone sequences were not available. The
individual or set of individuals as the aver- five other LCs were not included because of Phylogenetic Tree Analysis
age percentage of sequence divergence (ex- their slightly shorter nucleotide sequence
cluding positions with gaps) of all available
pairs of C2-V3 nucleotide sequences. Ge-
-
lengths. The pairwise distance measurements report-
As shown in Table 1, average distances ed in Table 1 do not provide information
netic distances between the viruses of the between the viruses of dental patients A, B, about either the relationship of the viruses
dentist and the patients (21), as well as the C, E, and G and those of the LCs ranged from the infected dental patients to each
ranges of differences, are shown in Table 1. from 10.8 to 11.8%. These distances are in other or the viral genetic distances from the
For the patients A, B, C, E, and G, the accord with the average interpatient dis- patients to individual LCs. Furthermore,
average distance to the dentist's viruses tance found among the 30 LCs, namely these measurements do not delineate the
clustered in the range of 3.4 to 4.9%, 12.0%. In addition, the average distances proximal source of the viruses infecting the
distances comparable to those reported for between the dentist's viruses and those of individual dental patients. To more closely
known epidemiologically linked infections the 30 LCs is 11.0%, whereas the average examine the full range of relationships be-
(2, 11, 12). In contrast, viruses present in distance between the dentist's viruses and tween the viruses of the dentist, the pa-
patients D and F were more distantly relat- those of patients A, B, C, E, and G is 4.0%. tients. and the LCs. we subiected the C2-
ed to those of the dentist, 13.6% and The possibility that the HIV sequences V3 nucleotide sequence data set to phylo-
10.7%, respectively. These latter distances in patients A, B, C, E, and G were closer in genetic tree analyses. These cluster analyses
are typical of equivalently measured dis- their genetic distance to the dentist's virus- were intended to be both "exploratory"
tances for viruses taken from epidemiologi- es than to the LC's viruses as a result of and, within the limitations of the sequence
callv unlinked infections and are consistent chance alone was tested using the Wil- lengths, statistical^ (24).
with the epidemiologicfinding that patients coxon rank-sum statistic. The test statistic A representative tree comprising the
D and F engaged in high-risk behaviors that was significant (P = 6 x suggesting dental group and select LC sequences is
could have resulted in their HIV infections.
These and other laboratory results (Fig. 1
and Table 2), in conjunction with the A
...................................
epidemiologic findings, led to the decision Patient A-y
to exclude patients D and F from the cluster Patient C-x
of linked infections in the statistical analy- Patient C-y
ses presented below. Patient G-x
When the percentages of sequence di- Patient A-y Patient G-y
vergence for viral sequences from the same
Patient G-x Patient A-x
person were calculated, the averages found Patient B-x
in the viruses of the dentist and each of the Patient G-y
Patient A-x Patient B-y
patients A, B, C, E, and G were low (0.0 to Patient E-x
5.4%). These data are consistent with pre- 7 . Patient B-x
Patient E-y
vious reports of intraperson genetic varia-
tion (2, 3, 6), indicating that none of the Patient E-x
five patients was infected with HIV from Patient E-y :Flg. 1. Phylogenetic tree analy-
....................................
more than one source. - LC2-x sis comparing HIV-1 envV3 re-
gion coding sequences from
Although no case of a stable form of LC3-x
HIV has been reported (3), the existence of the dentist,the dental patients A
LC2-y through G, and select local con-
a predominant HIV-1 variant in this part of
Florida could not be presumptively ruled
- Patient F-x trols LC2, LC3, LC9, and LC35.
Patient F-y The two most divergent clone
out. The existence of such a variant could
potentially explain the finding of highly
- LC Consensus Sequence sequences from each person
(designated x and y for each
LC9
related HIV sequences among patients A,
B, C, E, and G, independent of their - -
-
LC35
LC3-y
person), with the exception of
LC9 and LC35, for whom only
contact with the dentist. Therefore, we
patient D-x , 4% , direct PCR sequencing prod-
determined the HIV C2-V3 sequences of
the 35 HIV-seropositive LCs to examine
1 Patient D-y
ucts were available, have been
included. The LC sequences in-
the extent of genetic variation in the local cluded were those found by
geographic area. The sequences for 33 of pairwise distance measurement (Table 1) and signature pattern analysis (Table 2) to be the closest
the LCs were determined directly from the control sequences to the dental group sequences, which are enclosed by a box in (A). An LC
consensussequence has also been included,and the tree was rooted upon the African sample ELI.
amplified products (15 ) , which permitted The PAUP parsimony algorithm was used to analyze 279 aligned sites (25), of which 146 sites were
the determination of the most common varied. When the dentist's viral sequences were withdrawn from the analysis, or required to cluster
residue at each position (11, 22). In addi- with the LC consensus sequence (25), the dental clade remained otherwise unaffected, as shown
tion, the C2-V3 sequences from 7 of the 35 in (B).Vertical distances are for clarity only; the lengths of the horizontal branches are proportional
LCs, including the two for which direct to the single base changes and can be read as percentage differences with the scale bar.

SCIENCE VOL. 256 22 MAY 1992 1167

Table 2. Frequencies of amino acids that define a signature pattern in the dentist's viruses. Signature Pattern Analyses
Group Frequencies Early in the course of this investigation, we
noted the shared occurrence of a unique
Reference set*
E T E S T A I Q amino acid pattern in some clones of the
Reference set 0.63 0.81 0.69 0.59 0.69 0.66 0.72 0.56 dentist's and patient A's viruses (4). Al-
Florida LCs 0.67 0.67 0.63 0.75 0.42 0.60 0.84 0.45 though this particular pattern was not ob-
Dentist's viruses 0 0 0 0 0 0 0 0 served in all of the clones from these two
A, B, C, E, G viruses 0 0 0 0 0 0.20 0 0 individuals, nor in sequences from the oth-
D and F viruses 0.55 1.0 0.73 0.91 0 0.45 0.82 0.82
er dental patients, it encouraged us to
Dentist's signaturet
A I A G A E V H undertake a third measure of genetic simi-
Reference set 0.06 0.13 0.06 0.16 0.25 0.06 0.25 0.16 larity based upon phylogenetically informa-
Florida LCs 0.15 0.23 0.04 0.15 0.58 0.06 0.14 0.28 tive characters. This approach will be re-
Dentist's viruses 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 ferred to as amino acid signature pattern
A,B,C,E,Gviruses 1.0 0.94 0.97 1.0 1.0 0.69 1.0 1.0 analysis.
D and F viruses 0.36 0 0 0.09 1.0 0 0 0 A unique signature pattern of eight res-
*Thirty-twodistinct V3 region sequences available from the HIV Sequence Database (3). Those positions at which idues was detected within the 120-amino
the dentist'sviruses differed from that amino acid found in 50% or more of the reference set sequences constitute acid C2-V3 region of the dentist's viruses by
the dentist's signature pattern. Eight noncontiguous amino acids were objectively identified (see Fig. 2 for
positions involved). tThe dentist's signature pattern is found by definition in all six cloned sequences of the comparing an alignment of his viral amino
dentist's viruses. Signature patterns were also determined for patients A, B, C, E, and G. Signature similarity: acid sequences to an alignment of a refer-
dentist, 6 clones with 818 amino acids identicalwith signature pattern; patient A, 6 clones with 818; patient B, 11 ence set consisting of 26 North American,
clones with 818 and 1 clone with 718; patient C, 4 clones with 818 and 1 clone with 718; patient E, 4 clones with 818
and 2 clones with 718; patient G, 5 clones with 818; patient D, 4 clones with 218 and 1 clone with 118; and patient 5 Haitian, and 1 European HIV-1 se-
F, 1 clone with 218 and 5 clones with 118. quences (29). The resulting pattern consist-
ed of all sites in which all six clones from
the dentist differed from what is found in
shown in Fig. 1. A parsimony algorithm was LC consensus sequence in a "user-defined" over 50% of sequences in the reference set
used to generate this tree (25). Due to the tree (25), the remaining portion of the (Table 2 and marked with asterisks in Fig.
large number of LCs, not all clone sequences dental clade remained intact (Fig. 1B). 2; although consensus sequences are shown
available in this study could be satisfactorily Thus, tree analyses were consistent with for illustrative puiposes, every available
represented in a single tree. Therefore, we the genetic distance analysis summarized in clone sequence was used in this analysis).
focused the analysis on trees constituted Table 1, and the tree analyses defined a The dentist's signature pattern consisted of
from the two most divergent sequences strong linkage among the sequences from the following eight noncontiguous amino
(when available) from each of the samples patients A, B, C, E, and G, independent of acids: A, I, A, G, A, E, V, and H. In
and included only those LC sequences that the presence or absence of the dentist's viral contrast. the most common amino acids at
by distance measurement (Table 1) or signa- sequences. However, tree analyses could these sites in the reference set were E, T, E,
ture pattern analysis (Table 2) were found to not specify the order, that is, the direction, S, T, A, I, and Q, which occurred with
be most closely related to the dental group of transmission within the dental group. frequencies ranging from 0.56 to 0.81. No
sequences. In addition, a consensus se- It is prudent with cluster analyses in- sequence in the reference set contained
quence derived from all of the LC sequences volving a relatively small number of phylo- more than four of the dentist's signature
(LC consensus sequence) was included. genetically informative sites to subject the amino acids; most contained two or fewer.
Many equally parsimonious trees were found; cluster analyses to statistical bootstrap anal- We then inspected the viral sequences
however, the monophyletic nest containing ysis (27). In each of 100 iterations of the of the dental patients and the LCs for the
the sequences of the dentist and of patients bootstrap analysis, nucleotide sites were presence of the dentist's signature pattern.
A, B, C, E, and G was common to all of the randomly selected (with replacement) from No clone sequence from patients A, B, C,
most parsimonious trees. This nest, indicat- the 146 sites that had been used as the basis E, and G possessed fewer than seven of the
ed by the boxed area in Fig. lA, will be for construction of the phylogenetic tree eight signature amino acids (Table 2). As
referred to as the dental clade. shown in Fig. 1A. A new tree was then with the reference set of 32 sequences, most
Other trees were constructed by (i) using constructed for each iteration. By this pro- LC sequences showed two or fewer of the
consensus sequences; (ii) rearranging the cedure, the monophyletic grouping of the signature amino acids. In computing the
order in which sequences were presented to dental sequences was observed in 79 of 100 statistical significance of these findings, we
the computer program; (iii) including all replicates (26). The value 79 of 100 should used the sequences from those individual
LC sequences, other U.S.-derived se- not be construed as a P value for a statistical clones for each of the patients A, B, C, E,
quences, and V4-C3-V5 sequences; (iv) test of the hypothesis that the dental clade and G that agreed least with the dentist's
using only nucleotides representing the first forms a monophyletic nest; results of such signature. For the seven LC cases in which
and second base positions in codons; (v) tests (for instance, the genetic distance there were multiple clones, we used the
removing 64 noninformative sites (of the analyses that use rank-sum statistics and the clone sequence that agreed best with the
total 146 varied sites) (26); and (vi) succes- signature pattern analysis) are presented dentist's signature. This biased the outcome
sively removing and reintroducing the pa- below. Instead, the results should be inter- toward the hwothesis
L, that the viruses of
tient and LC sequences. All of the most preted as an indication of how often we the dental patients and the LCs were equal-
parsimonious trees generated in these dif- would conclude, from randomly selected ly similar to the dentist's viruses. Even so,
ferent ways retained the dental clade shown nucleotide sites, that the dental clade was a the Wilcoxon rank-sum test (P = 8 x
in Fig. 1; the viral sequences from patients distinct cluster. This procedure provides shows the five patients' viruses have
A, B, C, E, and G invariably nested with some measure of the power of the tree significantlybetter agreement with the den-
the dentist's viral sequences. When the analysis, although exact interpretation of tist's signature pattern than do the 28 LC
dentist's viral sequences were removed from results from bootstrap analysis of parsimony viruses for which agreement at all eight
the analysis, or required to cluster with the trees is a subject of controversy (28). signature amino acid sites could be assessed
1168 SCIENCE VOL. 256 22 MAY 1992
 (30). This finding corroborated the genetic viral sequence. No LC sequences or refer- Conclusion
distance analysis (Table 1) and phylogenet- ence set sequences were found to match as
ic tree analyses (Fig. 1). Furthermore, the many as 50% of the amino acids in the In summary, seven HIV-infected patients
viruses from patients D and F are again signatures from patients A, B, C, E, and G. were identified in the practice of a dentist
distinct from the viruses infecting the other Although the dentist was the most likely with AIDS. Of these seven patients, five
five patients; they possessed no more than source of the infection for the five patients, had no identified risk for HIV infection and
two signature amino acids. all the patients were not necessarily infected had invasive procedures performed by the
Both the genetic distance and signature with each of the dentist's HIV variants. This dentist. These five patients were infected
pattern analyses identified a few LCs that is evident upon inspection of the deduced with HIV strains that had nucleotide se-
were close to, but distinct from, the dental amino acid sequences, which demonstrate quences and amino acid signature patterns
cluster. LC35 and LC9 had C2-V3 nucleo- two C2-V3 subtypes for the dentist (Dent-I that were closely related to those of the
tide sequences that were closest to those of and Dent-I1 in Fig. 2). Both the dentist's dentist's viruses. In addition. the HIV se-
the dentist by genetic distance analysis, subtypes are also present in patient A's viral quences of the dentist and these five pa-
averaging 6.7% (range 5.8 to 7.8%) and sequences (Pt-A-I and Pt-A-11). However, tients were distinct from those found in the
6.9% (range 5.8 to 8.2%), respectively. sequences closer to Dent-I were found in two dental patients with known behavioral
The sequences from these two LCs pos- patients B and E, and sequences closer to risks for HIV infection and in 35 other
sessed four of the eight amino acids in the Dent-I1 were found in patients C and G. HIV-infected persons residing in the same
dentist's signature pattern (Fig. 1). The
maximum number of the dentist's signature
amino acids, five, found in any of the LC
sequences occurred in one clone sequence
from LC3. However, this clone had an
average distance of 9.8% from the dentist's
clone seauences. and the other five clones
from LC3 all contained only two signature
amino acids. By phylogenetic analyses,
these LC sequences did not appear to be
part of the dental clade (Fig. 1).
We next addressed the question of
whether two of the measures of genetic
similarity (genetic distance and amino acid
signature pattern analysis) were truly differ-
ent measures or were merely restatements of
the same measures. If the information con-
tained in these two measures overlapped,
we could expect to see a high degree of
correlation between them. However, Ken-
dall's T~ (31), a measure of correlation
computed from the 28 LCs for which both
measures were available, was -0.07 (95%
confidence interval from -0.37 to 0.24; a
value of 0 indicates a lack of correlation).
Therefore, we conclude that the analyses of
average distance and signature pattern are
separate assessments of genetic similarity.
With the discovery that phylogenetic
information is found in amino acid signa-
ture patterns, we also examined the signa-
ture patterns belonging to the dental pa-
tients' viruses. With the same reference set
and criteria employed in the definition of
the dentist's signature, unique signatures for
each of the dental patients' sets of viruses
were determined with the following results: Fig. 2. Comparison among V3 region amino acid sequences from 43 HIV-infected persons from
(i) patients A, B, and E have identical or Florida. For the purpose of this figure only, consensus sequences are shown. Because two distinct
nearly identical signatures to that of the sequence subtypeswere present in the dentist and patient A, two consensus sequences are shown
dentist and (ii) patients C and G have for the dentist (Dent-l and Dent-ll) and patient A (Pt-A-I and Pt-A-11).Sequences from the C2-V3
larger signatures, 15 and 14 residues, re- clone (n = 72) and direct (n = 41) sequences have been deposited with GenBank (accession
spectively (A, I, G, V, A, D, R, E, V, V, numbers M90847 to M90906;M90914 to M90966).Sequencesfrom fhe V4-C3-V5region have also
I,T,H,P,andVandA,I,A,G,A,D,V, been deposited with GenBank (accession numbers M91084 to M91115; M91121 to M91149;and
Y, E, V, V, I, H, and V). Nevertheless, the M91153to M91156).The reference sequence is Dent-l from the dentist. Residues identical to Dent-l
are shown as a dash. Regions not sequenced are left blank. The sequence stretch corresponding
signature from patient C's viruses is closer
to the V3 loop is indicated. Other symbols are ?, no consensus amino acid at the position indicated;
to the dentist's viral sequences than to any *, noncontiguous amino acids of the dentist's signature pattern (Table 2); -, a gap to maintain
other sequences. The signature from pa- alignment of the amino acids; and $, a stop codon. Single-letter abbreviations for the amino acid
tient G agrees best with a viral sequence residues are: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H , His; I, Ile; K, Lys; L, Leu; M, Met; N,
from patient C but next best to a dentist Asn; P, Pro; Q , Gln; R, Arg; S, Ser;T, Thr; V, Val; W, Trp; and Y, Tyr.
SCIENCE VOL. 256 22 MAY 1992 1169
geographic area. Therefore, both the epide- netic variation and analysis of greater se- containing a 9-base extension (5'-CTAGCAGAA-
3') at the 3' end of CL207. Both the CL207 and its
miologic and laboratory data indicate that quence lengths may be necessary to identify extended derivatives have an Eco RI restriction
these five patients were infected during strain variation and to determine possible endonuclease recognition sequence (GAATTC).
their dental care. epidemiologic linkage. Nonetheless, this Amplified DNA from the secondary PCR was
purified by NACS columns (Bethesda Research
The precise mode of HIV transmission to investigation demonstrates that detailed Laboratory, Gaithersburg, MD), sequenced di-
these patients could not, however, be iden- analysis of HIV genetic variation is a new rectly or digested with Eco RI, and cloned into
tified. Patients could have been directlv and powerful tool for understanding the MI3 vectors. To assist the tracking of recombi-
exposed to the dentist's blood as a result of epidemiology of HIV transmission. nant phages generated from the dentist and the
patients A through G, a specific dinucleotide
percutaneous injuries sustained by the den- Note added in @roof: Subsequent to the sequence was added 3' to the Eco RI site of each
tist while performing invasive procedures on preparation of this report, the Centers for of the CEO7 derivatives to serve as person-
these patients. Although none of the pa- Disease Control became aware of another specific dinucleotides or "bar codes." When pos-
sible, a different MI3 vector was used to generate
tients reported that the dentist injured him- HIV-infected dental patient (patient H). clones from different persons in the dental cohort.
self while caring for them, they would not This patient has acknowledged behavioral These two additional measures facilitated the
necessarily have been aware if such injuries risk factors for HIV infection; his HIV clone management and identification of the source of
HIV-containingphage. MI3 clones containing HIV
had occurred. Transmission might also have sequences (GenBank accession numbers inserts were screened by plaque hybridization
occurred by contamination of instruments or M90907 through M909 12) are not closely using 32P-labeled C0249 (5'-CAAATATACA-
other dental equipment with blood from the related to the dentist's viruses, differing by GGGCTGCATTAACAAGAGATGGTGGTAA-3') or
its complementary sequence C0250 (5'-TAC-
dentist or, less likely, a patient or patients more than 10.7% in the C2-V3 region. CACCATCTCTTGTTAATAGCAGCCCTGTAA-
already infected by the dentist. All five TAmG-3') as probes derived from the con-
patients had invasive procedures performed REFERENCESANDNOTES served C3 domain.The Taq Dye Primer Sequenc-
ing Kit and the 373A DNA sequencer (Applied
by the dentist after he was diagnosed with 1. J. S. Smith et ab, N. Engl. J. Med. 324, 205 (1991); Biosystem, Foster City, CA) were used. Fluores-
AIDS. At this late stage of disease, higher H. J. Lin etab, J. Infect.Dis. 164, 284 (1991); G. P. cence-tagged MI3 universal primers were used
virus titers may have been present in the Holmes et a/.,Ann. Intern. Med. 112, 833 (1990); for sequencing M13-derived clones, whereas a
R. Rico-Hesse, M. A. Pallansch, B. K. Nottay, 0. flu~rescence-taggedderivative of CEO7 primer
dentist's blood and he may have been more was used for direct sequencing of amplified DNA.
M. Kew, Virology 160, 311 (1987).
likely to transmit virus than earlier in the 2. P. Balfe et a/., J. Virol. 64, 6221 (1990). 16. M. Rogers et a/., N. Engl. J. Med. 320, 1649
course of his HIV disease (32). 3. G. Myers et a/., Eds., "Human retroviruses and (1989).
The theoretical ~ossibilitvof HIV trans- AlDS 1991: A compilation and analysis of nucleic 17. Precautions included (i) ultraviolet irradiation of
acid and amino acid sequences," (Theoretical vials and laboratory benches ( l a ) ,(ii) apportion-
mission from infected health-care workers to Division, T10, Los Alamos National Laboratory, ing reagents for single use, (iii) use of positive
patients during invasive procedures has been Los Alamos, NM, 1991). displacement pipetting devices, and (iv) use of a
previously acknowledged (33), and there are 4. Morbid. Mortal. Wkly Rep. 39, 489 (1990). room dedicated specificallyfor PCR (19).
5. Ibid.40, 21 (1991); ibid., p. 377; C.A. Ciesielski et 18. G. Sarkar and S. S. Sommer, Nature 343, 27
well-documented reuorts of HIV transmis- a/., Ann. Int. Med. 116, 798 (1992). (1990); C.-Y. Ou, J. L. Moore, G. Schochetman,
sion from patients to health-care workers 6. J. P. Vartanian, A. Meyerhans, B. Asjo, S. Wain- BioTechniques 10, 442 (1991).
after percutaneous exposure to HIV-infected Hobson, J. Virol. 65, 1779 (1991); S. Delassus, R. 19. S. Kwok,Amplifications2, 4 (1989); a n d R.
blood (34). In addition, transmission of Cheynier, S. Wain-Hobson, ibid., p. 225; A. Myer- Higuchi, Nature 339, 237 (1989).
hans etal., Cell58, 901 (1989); M. Goodenow et 20. The V3, V4, and V5 sequences of patients A and
hepatitis B virus, which has epidemiologic a/., J. Acquired Immune Defic. Syndrome 2, 344 B and a local control (GenBank accession num-
transmission patterns similar to HIV, from (1989); M. Alizon etal., Cell46, 63 (1986); B. R. bers M92100 to M92150) were also determined
health-care workers to patients during inva- Starcich et a/., ibid. 45, 637 (1986). by A. J. L. Brown and his colleagues at the
7. B. H. Hahn etal., Science232, 1548 (1986); M. S. University of Edinburgh. PBMC DNA from patient
sive medical and dental ~rocedureshas been Saag et a/., Nature 334, 440 (1988). A and an LC were prepared in a CDC laboratory
reported (35). For most of these outbreaks, 8. J. A. McKeating et a/., AlDS 3, 777 (1989); P. in which no work had been done with HIV. PBMCs
the ~recisemode of he~atitisB virus trans- Nara, in Retrovirusesof HumanAlDS and Related from patient B were sent directly from CDC's
Animal Diseases, M. Girard and L. Valette, Eds. Epidemic Response Laboratory to Edinburgh.
mission from the worker to patient could not (Pasteur Vaccins, Paris, 1989), pp. 203-215. None of the three specimens were handled by
be identified, as was true for the HIV trans- 9. T. Shioda, J. A. Levy, C. Cheng-Mayer, Nature CDC personnelinvolved in the PCR and sequenc-
missions reported here. 349, 167 (1991). ing of the Florida specimens.
Analysis of viral nucleotide sequence 10. C. A. B. Boucher etal., Lancet336,585 (1990); B. 21. For the nucleotide sequence comparisons shown
A. Larder, G. Darby, D. D. Richman, Science243, in Table 1, a total of 94 sequences, derived from
distances, inferred phylogenetic relation- 1731 (1989); D. D. Richman, J. M. Grimes, S. W. both MI3 clones and from direct sequencing of
ships, and signature patterns of deduced Lagakos, J. Acquired Immune Defic. Syndrome3, amplified products, were analyzed over 243
amino acid sequences played a central role 743 (1990). shared positions in the C2-V3 region, for a total of
11. H. Burger et a/., Lancet 336, 134 (1990); H. 4371 pairwise comparisons. In keeping with cur-
in the investigation of this cluster of HIV- Burger et a/,, Proc. Natl. Acad. Sci. U.S.A. 88, rent practice, averages and ranges of sequence
infected ~atients. Similar analvses mav 11236 (1991). differences are reported as percentages rather
prove to be useful in other investigationsof 12. S. M. Wolinskv etal., Science 255, 1134 (1992). than as number of nucleotide differences. The
13. A. Srinivasan k t a/., Blood 232, 1548 (1986). robustness of this analysis, although involving
HIV transmission, such as those seeking to small differences in nucleotide counts, is support-
14. K. Cichutek et al., AlDS 5, 1185 (1991).
determine the source of HIV infection in 15. Peripheral blood mononuclear cells (PBMCs) ed by (i) the outcome of the Wilcoxon rank-sum
persons with multiple risk factors (for ex- were isolated, divided, and processed in a labo- analysis; (ii) the consistency observed for the
ratory where no PCR work had been performed. ranges, which do not show marked variation (Ta-
ample, a health-care worker who has sus- ble 1); and (iii) the agreement of the intrapatient
To avoid potential cross-contaminationor mix-up
tained an occupational exposure to HIV of specimens, only one frozen PBMC aliquot of differences with what has been reported (2, 3, 6).
and reports male-to-male sexual contact). the dentist or his patients was processed at a 22. H. Steuler, B. Storch-Hagenlocher,B.Wildemann,
~ e n e t i canalysis is also proving to be valu- time. Amplification of HIV env DNA was per- AlDS Res. Hum. Retroviruses 8, 53 (1992); R.
formed by a two-step PCR procedure. The primer Gibbs et a/., in PCR Technology: Principles and
able in tracking the spread of HIV in pair used in the first amplification step was CL207 Applications for DNA Amplification (Stockton,
countries, such as Thailand, where the (5'-GTATGAATCAACTGCTGTAAATGGCAGT- New York, 1989), pp. 171-191
e~idemicis of recent onset (36). . , 3') and C072 (5'-TATAGAATCACTCTCCAAT-
TGTCCCTCAT-3').Approximately 1 pg of PBMC
23. The Wilcoxon rank-sum statistic was calculated
from the average distance between the dentist
In the current investigation, the diver- DNA was used in a standard 100-pl reaction (16) and each of patients A, B, C, E, and G and the 30
gence of HIV sequences within the Florida and the PCR profile was 96"C, 1 min, 65"C, 2.5 LCs for whom genetic distance measurements
background population was sufficient to min, temperature ramping time 30 sand the cycle were available. When clone sequences were
was repeated 40 times. Five microliters of the available, these were used in preference to direct
identify strain variation. However, in coun- amplified product was diluted 50-fold and 5 pI of sequences. In all Wilcoxon statistics reported
tries where the virus has been recently the diluted DNA was reamplified 30 cycles with a here, average ranks were used in case of ties. All
introduced, there may be limited HIV ge- new primer pair C072 and derivatives of CL207 P values for the Wilcoxon statistics given in this

SCIENCE VOL. 256 22 MAY 1992

 paper were obtained from the exact distributionof the bootstrapping turned out to be negligible the Pvalue, which here is simply twice the recip-
the rank-sum statistic [for example, see T. P. (80% versus 79%). rocal of the binomial coefficient (y),may be
Hettmansperger Statistical Inference Based on 27. Bootstrap analysis utilized the DNABOOT pro- questioned. Thus, we also conducted an analysis
Ranks(Wiley, New York, 1984)l.Since the Pvalue gram of PHYLIP (24). Increasing the number of with only direct sequences from the dental pa-
forthe Wilcoxon statistic may depend on the mix iterations, or replicates, from 100 to 400 yielded tients and the 33 LCs for whom direct sequences
of direct and clone sequences, we also conduct- the same proportion of trees in which the dental were available. Because the direct sequences of
ed a bootstrap hypothesis test [P. Hall and S. R. clade could be identified.There is no consensus two of the dental patients were too short to assess
Wilson, Biometries 47, 757 (1991)l.This analysis about the interpretation of bootstrap results; on agreement at all eight sites, we tested the hypoth-
also took into account the varying number of the one hand, confidence intervals may be larger esis that the probability of agreement at isites,
clone sequences available per individual. We than the set of equally parsimonious trees (24) given that jwere available, was equal among the
found a 95% confidenceintervalfor the Pvalue to but, on the other hand, bootstrap intervals can be dental patients and the LC by Fisher's exact test
be between 2.7 x 10-6 and 4.4 x 1OW6. We also underestimatesof confidence intervals (28). (P= 1.1 x [G. H. Freemanand J. H. Halton,
computed a Wilcoxon rank-sum statistic from av- 28. D. Hillis and J. Bull, University of Texas, Austin, Biometrika38, 141 (1951)l.
erage distances between the cloned viral se- unpublished observations. Becauseagreement in the signature pattern anal-
quences of the dentist and the direct sequences 29. The choice of 32 reference sequences was dic- ysis is measured on a scale from 0 to 8, we have
of the five dental patients and those 28 LCs for tated solely by what was available for analysis in used a form of Kendall's T, denoted T,, which is
whom direct sequences of sufficient length were the HIV Sequence Database, with the following adjusted for ties [A. Stuart, in Encyclopedia of
available (P= 8 x Finally, we computed a qualifications: (i) sibling sequences (sets of viral StatisticalSciences (Wiley, New York, 1983), vol.
Wilcoxon rank-sum statistic from distances from sequences from the same patient) were not in- 4, p. 3671. A less familiar measure, which uses
the direct sequence of the dentist to the direct cluded; (ii) only non-African, "MN-like" se- only the untied data and hence may be a better
sequencesfrom the patient group and the 28 LCs quences [G. J. LaRosa et ab, Science 251, 811 estimator when data have extensive ties, is
(P = 2 x 10-5). (1991)l were included in light of the fact that none Kruskal and Goodman's y. We find that y = 0.08,
24. J. Felsenstein, Evolution 39, 783 (1985); Annu. of the dental or Florida control sequences mani- with asymptotic confidenceinterval (-0.42, 0.27).
Rev. Genet. 22, 521 (1988). fested "African-like" V3 region sequences; and D. D. Ho, T. Moudgil, M. Alam, N. Engl. J. Med.
25. Tree analyses were principallyconductedwith the (iii) whereas over 1000 "V3 loop" sequences (33 321, 1625 (1989); R. W. Coombs et a/., ibid., p
use of the MULPARS and branch and bound to 35 amino acids) are present in the HIV data- 1626.
programs of PAUP (PhylogeneticAnalysis Using base, the current analysis required complete V3 M i b i d . Mortal. Wkly. Rep. 36, 2s (1987); ibid. 35,
Parsimony, version 2.4.2, written and distributed region sequences. The HIV-1 reference set com- 221 (1986).
by D. L. Swofford, Illinois Natural History Survey, prises MN, M I , JRCSF, A M I , ADA, RF, WMJ2, R. Marcus and the CDC Cooperative Needlestick
Urbana). The LC consensus sequence was the JH3, SC, BRVA, BALI, JFL, OYI, NY5, SF162, Surveillance Group, N. Engl. J. Med. 319, 1118
designated hypothetical ancestor (Hypanc). SF2, SF33, HAN, CDC451, SBA, MA221, JB02, (1988).
Trees were rooted upon the African sample se- FJS, WM, ACH9, JM, ACPI, 6008ar, 5986ar, J. Welch et aL, Lancet i, 205 (1989); F. E. Shaw,
quence ELI or upon the midpoint of the greatest 6000ar, 5998ar, and 6002ar (3). Signature amino Jr. etal., J. Am. Med. Assoc. 255,3260 (1988); M.
patrisitic distance with the same outcome. In a acids in the dentist's viruses are those that differ A. Kane and L. A. Lettau, J.Am. Dent.Assoc. 110,
"user-defined"tree, the dentist'ssequenceswere from what is found in 50% or more of these 834 (1985); J. Ahtone and R. A. Goodman, ibid.
required to cluster with the LC consensus se- sequences. Reduction of the reference set by 106, 219 (1983).
quence. To examine the consequences of rear- exclusion of the last five sequences in the set (the C-Y. Ou et ab, AIDS-Res. Hum. Retroviruses, in
ranging the order of input of sequences, trees Haitian samples, kindly provided by N. Halsey press.
were constructed using ,the JUMBLE option of and his colleagues and sequenced at CDC) R. F. Smith and T. F. Smith, Protein Eng. 5, 35
DNAPARS program of PHYLIP (version 3.2, writ- made no difference in the determination of this (1992)
ten and distributedby J. Felsenstein,Universityof signature. Signature analysis utilizing nucleotide We thank A. Gifford and R. Lenroot for providing
Washington, Seattle). sequences rather than amino acid sequences excellent assistance in computer analysis; A. L.
26. Of the 146 varied sites, 64 sites manifested nu- confirmed without exception the results reported Brown and his colleagues for independently con-
cleotide changes in no more than one of the here. A fuller description of the computer appli- firming the HIV sequences in patients A and B
sequences included in the analysis; for the pur- cation (VESPA, or viral epidemiology signature and an LC; D. K. Hillis for his assistance with the
poses of some bootstrapping evaluations, these pattern analysis) and the results with nucleotide phylogenetic tree analysis; B. Holloway, E.
s,ites were deemed noninformative [W.-M. Li and signature patterns is being prepared (B. T. M. George, and N. de la Torre for synthesis and
D. Graur, in Fundamentalsof Molecular Evolution Korber and G. Myers, unpublished observations). purificationof DNA oligomers; D. Hammerlein for
(Sinauer, Sunderland, MA, 1991), pp. 111-1 131 30. The Wilcoxon rank-sum statistic was calculated handling blood specimens; and R. Moseley for
and were accordingly excluded. The effect upon from a mix of clone and direct sequences. Hence excellent editorial assistance.

SCIENCE VOL. 256 22 MAY 1992